The Google Ads Tactic That Cut Invalid Clicks by 50% (and How to Copy It)

Google Said It Caught Everything. The Clicks Kept Coming

On July 2, 2026, Search Engine Land published a case study with an uncomfortable premise. A client of StubGroup, a Google Premier Partner agency led by John Horn, was watching relevant, high-intent search traffic refuse to convert. Google's own reporting put the account's invalid click rate at 60% to 80%. The agency filed an investigation. Google agreed the activity looked suspicious, said its filters had already caught all of it, and had not charged for it.

The traffic still would not convert.

So the agency stopped waiting. It added 540 of Google's predefined audiences to the client's Search campaigns, set to Targeting instead of Observation. Reported invalid clicks immediately fell by 50%, and conversion rates climbed back to profitable levels. This post covers what the case found, why Google's automatic filtering and the usual IP blockers were not enough, the exact steps to copy the setup, and who should leave it alone.

The case: 540 audiences, half the invalid clicks

The client sold book editing and ghostwriting services, a niche with heavy competition and expensive clicks. The search terms triggering the ads were exactly right. The results were not. That gap between relevant traffic and zero conversions is what started the investigation.

Four signals that pointed to click fraud

Before touching anything, the team documented the evidence. Per the case study, four signals stood out:

• Google itself reporting a 60% to 80% invalid click rate on the account.
• Microsoft Clarity session recordings showing bot-like behavior from Google Ads traffic.
• Click-through rates above 80% across numerous search terms, with some exceeding 100%.
• Far fewer sessions in GA4 and other analytics tools than the number of clicks Google Ads reported.

Each signal alone is ambiguous. Together they describe traffic that clicks like a mob and behaves like nobody. A CTR over 100% is not a great ad, it is an accounting impossibility for genuine users, which is worth remembering next time a dashboard number looks too good: a high CTR does not mean your ads work.

The escalation that went nowhere

The team first tried third-party click fraud tools and, in the case study's words, saw no measurable performance improvement. Then it filed a formal investigation with Google. Google's answer was the frustrating kind of reassurance: yes, there was suspicious activity, and no, there was nothing to fix, because its systems had caught it all and had not billed for it.

The agency did not believe the filtering was complete. The conversion data sided with the agency.

The fix that finally moved the number

The change was one setting applied at scale. The team added 540 Google-defined audiences to the Search campaigns, deliberately choosing the Targeting setting rather than Observation, and deliberately not limiting the list to audiences that matched the client's buyer persona. The audiences were not there to describe the customer. They were there as a filter for accounts that look like people at all. Invalid clicks dropped by half, and the campaigns went back to making money.

What invalid clicks are (and what Google already does about them)

To copy the tactic responsibly, it helps to be precise about the problem. Invalid clicks are not one thing, and Google is not asleep at the wheel. The gap the case exposes is narrower and more interesting than "click fraud is rampant."

Google's definition, in Google's words

Google's help center defines invalid clicks as "Clicks on ads that aren't the result of genuine user interest, including intentionally fraudulent traffic and accidental or duplicate clicks." The examples Google gives span the malicious and the mundane:

• Manual clicks intended to inflate your costs or profit the sites hosting your ads.
• Clicks by automated clicking tools, robots, or other deceptive software.
• Accidental clicks with no value, like the second click of a double-click.

The malicious end of that spectrum is expensive at industry scale. A projection cited in the case study estimates advertisers will lose $172 billion a year to ad fraud by 2028.

What you pay for, and what comes back

Google's position is clear: "You won't be charged for invalid clicks or impressions as they provide little or no value." Detection works in two passes:

• Filtered live: clicks caught in real time never bill at all.
• Credited later: clicks identified as invalid after the fact come back as a credit toward future advertising, visible as invalid activity credit in your account.

The catch is structural. Refunds only exist for clicks Google flags. A fraudulent click that slips past detection is billed like a legitimate one, converts like a ghost, and quietly poisons every downstream metric you optimize against.

Where to see your own numbers

You do not have to guess whether this affects you. Three places to look:

1. The Invalid activity credit report in Report Editor, which the case study calls the most detailed view available.
2. The Invalid clicks and Invalid click rate columns, which you can add at the campaign level (not at the ad group or keyword level).
3. Your analytics: compare GA4 sessions against billed Google Ads clicks and investigate any persistent gap.

For calibration, a February study cited in the case found an 11.4% invalid click rate across 43,700 accounts, while the agency reports competitive industries running above 40%. Knowing your CTR baseline per search term helps too, because per-term CTR spikes are the earliest visible symptom.

Why the usual defenses fell short

The case is interesting precisely because the client did everything the standard playbook says. The playbook has three layers, and each one has a structural hole.

Google's filter is good, not complete

Automatic filtering catches a lot, and the credits are real. But the case documents the failure mode nobody controls: Google reviewing the account, agreeing the activity is suspicious, declaring it fully handled, and being wrong about the remainder. There is no appeal lane for "your filter missed some." You cannot see inside the detection system, so when filtered-plus-billed traffic still refuses to convert, your only move is to change what traffic reaches the auction in the first place.

The 500-IP ceiling meets infinite addresses

The classic self-serve defense is blocking bad IP addresses. Google's own documentation caps that lever: "You can exclude up to 500 IP addresses per campaign." Fraud operations are not built out of 500 addresses. As the case study puts it, fraudsters know how these tools work and often cycle through IP addresses using VPNs to stay one step ahead of the monitoring software. A blocklist is a photograph of yesterday's attack.

Third-party blockers inherit the same weakness

Click-fraud tools automate that same photograph. They watch for suspicious behavior, then exclude the addresses it came from, inside the same 500-slot budget. Against a static attacker that works. Against address rotation it decays instantly, which matches the client's experience: the tools produced no measurable improvement. Blocking where the fraud came from keeps losing to filtering who the click pretends to be. That reframing is the actual insight of the case.

How the audience filter works

The tactic sounds almost too simple: add hundreds of audiences to a Search campaign. The mechanism underneath is worth understanding before you copy it, because the same mechanism explains who should not.

Google's predefined audiences, repurposed as a sieve

Google maintains hundreds of prebuilt audience segments based on demographics, search behavior, and browsing behavior. Research private jets and luxury watches for a week and Google files you among luxury shoppers. Membership is not something a visitor claims. It is something Google infers from a long, expensive-to-fake trail of activity, the same signal layer advertisers normally use to target audiences on any platform.

That trail is the point. A real buyer of ghostwriting services has months of searches, sites, and videos behind their profile. A bot spun up behind a fresh VPN address has none.

One dropdown does all the work

When you attach audiences to a campaign, Google offers two settings, and they are opposites. Per Google's documentation, "Targeting restricts the reach of your ad group": your ads only show to people who trigger your keywords and belong to a selected audience. Observation is the passive twin. In Google's words, when you use the Observation setting, "the reach of your campaign or ad group isn't affected." It reports how audience members perform but keeps showing ads to everyone.

The team chose Targeting on purpose. Observation would have produced a nice report about the fraud while continuing to pay for it.

Why bots flunk the audience test

The agency's hypothesis, stated plainly in the case: fraudsters cycling through IP addresses are not always taking the time to build normal-looking online profiles that fit Google's predefined audiences. Requiring membership in any one of 540 segments is a laughably low bar for a human. Nearly everyone with a browsing history qualifies for something. It is a surprisingly high bar for a disposable bot identity. Audience membership works as the cheapest proof-of-humanity check in Google Ads: not because any single segment is precise, but because having no segment at all is the tell.

Copy the setup: the checklist

The change itself takes minutes. The discipline around it is what makes the result readable. Three phases: baseline, apply, verify.

Phase 1: baseline your evidence

Do this before changing anything, or you will never know whether it worked.

1. Add the Invalid clicks and Invalid click rate columns to your campaign view and record the current rate.
2. Pull the Invalid activity credit report in Report Editor for the last 90 days.
3. Note CTR per search term and flag anything above 30%, and anything above 100% twice.
4. Compare GA4 sessions against billed clicks for the same period and record the gap.
5. Record conversion rate and ROAS so profitability has a before-and-after.
6. Optional but persuasive: run session recordings (the case used Microsoft Clarity) on ad traffic for a week.

Phase 2: apply the audiences

The mechanical steps, straight from the case study:

1. Open the affected Search campaign.
2. Select Audiences, then Edit audience segments.
3. Choose Targeting, not Observation. The setting is the mechanism.
4. Click Browse and add audiences in bulk. Do not curate for persona fit; the case added 540 segments spanning most of what Google offers, because the goal is filtering profile-less traffic, not describing the customer.
5. Save, and change nothing else in the campaign that week. One variable at a time, the same discipline you would apply to building the campaign in the first place.

Phase 3: read the results

Give it two weeks of normal spend, then compare against your baseline. Success looks like the case's outcome:

• Invalid click rate meaningfully down (the case saw 50% immediately).
• Per-term CTRs falling back to plausible single and low double digits.
• The GA4-sessions-versus-clicks gap narrowing.
• Conversion rate and ROAS recovering as billed clicks turn back into humans.

If invalid clicks drop but conversions do not recover, your problem was never only fraud, and the next suspect is the offer or the landing page, not the traffic.

Who should not copy this

An honest reading of the case includes its own warning label. The agency states it only recommends this for accounts with high invalid click rates. Several kinds of accounts should skip it entirely.

If your invalid rate is normal, this is not for you

Against the study's 11.4% average, Google's default handling is doing its job: filtered, credited, done. The tactic exists for the pathological tail, accounts sitting far above average in competitive, high-CPC niches. Deploying it on a healthy account is treatment without a diagnosis.

The reach you give up is real

Google recommends the Observation setting for all Search campaigns, and the case study openly flags the downside of overriding that: you can unintentionally block legitimate users who do not fit within Google's predefined audiences. Some real buyers browse in privacy modes, decline personalization, or simply have thin profiles. With Targeting on, they never see your ad. For a broad account with modest fraud, that trade is a net loss.

Use it or skip it: the short version

Use it when:

• Google's own columns report an invalid click rate far above the ~11% average.
• CTRs per term exceed 80% or breach 100%.
• Analytics sessions run far below billed clicks, and recordings look robotic.
• Third-party blockers and a Google investigation both failed to restore conversions.

Skip it when:

• Your invalid click rate is near average and credits are flowing normally.
• Your niche is small and every impression of reach matters.
• Your buyers plausibly live outside profiled audiences, such as privacy-conscious segments.
• You have not baselined anything yet, since you could not measure the effect anyway.

What cleaner traffic buys you: the budget math

Filtering junk clicks is not the end goal. It is a way to get your budget and your data working again, and the second effect is bigger than the first.

A worked example

Say a competitive-niche account spends $3,000 a month on Search, and its true junk share is 30% of billed clicks, with Google's filters catching two-thirds of that. Run the numbers:

• 30% junk share, two-thirds caught by Google, means 10% of billed clicks slip through unflagged.
• 10% of $3,000 is $300 a month billed for clicks that can never convert.
• Annualized, that is $3,600 in pure waste, before counting the quieter cost: every automated bid strategy learning from polluted conversion data.

Cut the ghost share in half and you recover real spend while your cost per click starts buying measurable humans again.

Reinvest where the leverage is: creative volume

Recovered budget has a best use, and it is not more of the same keywords. Once traffic quality is fixed, creative becomes the binding constraint, especially as Google keeps reshaping the ad surface itself, wrapping AI summaries around your ads. At roughly $2 to $11 per AI-generated video clip, that recovered $300 a month funds somewhere between 27 and 150 fresh ad variants, enough to test angles continuously instead of betting a quarter on one asset.

How Novoads solves the creative side

Novoads is a global AI UGC video-ad generator: write or auto-generate a script, pick an AI actor, and get a UGC-style video ad in minutes, in 29+ languages with real regional accents. It is the volume half of this equation, turning reclaimed ad spend into a steady stream of testable creative instead of a single expensive shoot. Start for $1: the trial costs $1 for 3 days of access, then continues at $49 per month, and you can cancel anytime.

Clean traffic is a targeting decision

The lasting idea in this case is not the 50%. It is that traffic quality turned out to be something an advertiser could act on directly, with a native campaign setting, after every dedicated anti-fraud layer had shrugged. Fraudsters can fake a click and rotate an address. What they have not bothered to fake, so far, is a life online, and Google's audience system is a ready-made test for exactly that.

If your account shows the four signals, run the checklist and let your own before-and-after decide. Bots are not the only budget leak worth auditing either: competitors buying your brand name drain accounts just as quietly. Plug both, then spend what you reclaim on the thing junk traffic was starving: enough real creative to find your next winner.